Re: graceful restarts & log file rotation again

["Roy T. Fielding" <fi...@kiwi.ics.uci.edu>]

>Perhaps there's a USR1 bug in 1.3 now. 

If the timeouts are now scoreboard driven, the old children may be
losing their timeout constraint on a USR1 and spinning forever.
OTOH, I can't remember right now whether or not Dean's timeout changes
were committed to 1.3.  Hmmm, it says so in cvs log http_main.c,
but not in CHANGES.  Brian, did you compile with -DOPTIMIZE_TIMEOUTS ?

....Roy

Re: graceful restarts & log file rotation again

[Dean Gaudet <dg...@arctic.org>]

OPTIMIZE_TIMEOUTS is the default for anything with a memory scoreboard. 

Interesting idea though, I'll look for it.  But I don't think this is it
... since the parent scans the entire scoreboard (well, up to the last
non-DEAD slot in my upcoming cleanup patch) and does timeout processing on
all non-DEAD slots. 

Dean

p.s. re: CHANGES ... I thought I had a generic "reduced syscall overhead" 
comment in there.  I guess I need to add one.

On Mon, 8 Sep 1997, Roy T. Fielding wrote:

> >Perhaps there's a USR1 bug in 1.3 now. 
> 
> If the timeouts are now scoreboard driven, the old children may be
> losing their timeout constraint on a USR1 and spinning forever.
> OTOH, I can't remember right now whether or not Dean's timeout changes
> were committed to 1.3.  Hmmm, it says so in cvs log http_main.c,
> but not in CHANGES.  Brian, did you compile with -DOPTIMIZE_TIMEOUTS ?
> 
> ....Roy
> 

Re: graceful restarts & log file rotation again

[Brian Behlendorf <br...@organic.com>]

At 05:43 PM 9/8/97 -0700, you wrote:
>>Perhaps there's a USR1 bug in 1.3 now. 
>
>If the timeouts are now scoreboard driven, the old children may be
>losing their timeout constraint on a USR1 and spinning forever.
>OTOH, I can't remember right now whether or not Dean's timeout changes
>were committed to 1.3.  Hmmm, it says so in cvs log http_main.c,
>but not in CHANGES.  Brian, did you compile with -DOPTIMIZE_TIMEOUTS ?

It looks like httpd.h will set it, on BSDI at least.

#if !defined (MULTITHREAD) && (defined (HAVE_MMAP) || defined (HAVE_SHMGET))
#define OPTIMIZE_TIMEOUTS
#endif

Let me recompile with that undefined and see if it happens tonight...

	Brian


--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=--
"Why not?" - TL           brian@organic.com - hyperreal.org - apache.org